Writprint
Specimen
Not for filing
Sample ReportFile DEMO/A-001

A flagged essay in first-year composition

An English 101 instructor flagged a final argumentative essay as stylistically inconsistent with the same student's earlier coursework. The student denies the allegation. Three classmates with access to the prompt and a habit of sharing drafts are under consideration as possible ghostwriters.

View workplace investigation sample
Inputs3 candidates
01Student A

"I think the policy on AI tools is too vague right now. Some of my professors say it's fine to use it for brainstorming and others act like even opening ChatGPT is cheating. I've gotten different feedback on basically the same kind of work in two different classes. It would help if the school just picked a position and stuck to it."

02Student B

"Universities ought to adopt a clear and consistent policy on the use of artificial intelligence in academic work. Furthermore, such a policy must be enforced uniformly across departments. The current situation, in which different professors apply different standards, creates confusion among students and undermines the credibility of the institution as a whole."

03Student C

"AI tools should not be allowed for graded work. If you don't write the paper, you didn't do the assignment. The rules are simple. Either it's your work or it isn't. Schools that let people use these tools are basically letting students take credit for things they didn't do, and that's not fair to the students who actually did the work."

Target document (anonymous)

"The introduction of artificial intelligence into the undergraduate classroom requires a coordinated institutional response. Furthermore, such a response must address not only the immediate question of whether these tools are to be permitted, but also the broader question of how academic integrity is to be preserved in an environment where the conventional boundaries of authorship have become increasingly unclear. Universities that fail to articulate a coherent position on this issue will find themselves managing a sequence of ad hoc disputes rather than enforcing a defensible policy. The longer the matter is deferred, the more difficult it becomes to address."

Determination
Student B
Confidence
medium
Method
Closed-set stylometric attribution
with LLM-assisted feature extraction
Executive summary

The target essay most closely matches Student B's writing on register, sentence structure, and transitional vocabulary. The signal is meaningful but not conclusive: candidate samples are short (~55 words each), and the target is more polished than Student B's known work, raising the possibility of LLM-assisted editing rather than ghostwriting by a classmate. We recommend collecting a longer in-class writing sample under controlled conditions before any disciplinary action.

Ranked Attribution3 candidates
01

Student B

Primary attribution
71%

The target document is most consistent with Student B's writing on three independent dimensions. Both share a formal register marked by passive constructions ("is to be permitted," "must be enforced"), the same characteristic transition ("Furthermore,") in the second sentence, and a preference for noun-phrase modification ("the conventional boundaries of authorship," "a coherent position on this issue") over verb-driven prose. Both also avoid first-person pronouns and contractions entirely, in contrast to Students A and C. The match is meaningful but not conclusive — the candidate sample is short, and the target essay is more polished than even Student B's strongest known work.

Stylistic Evidence
  • ›Sentence-initial "Furthermore," appears in both target and Student B's sample — absent in A and C
  • ›Passive voice constructions per 100 words: target ≈ 4.5, Student B ≈ 3.9, Students A and C ≈ 0
  • ›First-person pronouns: 0 in target, 0 in Student B's sample, 4 in Student A's sample, 1 in Student C's sample
  • ›Contractions: 0 in target, 0 in Student B's sample; present in both Student A ("it's," "I've") and Student C ("don't," "didn't," "that's")
  • ›Mean sentence length: target ≈ 28 words, Student B ≈ 27, Student A ≈ 19, Student C ≈ 14
  • ›Both target and Student B use "in which" as a relative-clause construction — absent in A and C
02

Student A

15%

Student A's sample shares the topical concern with policy ambiguity, but the register is consistently informal — first-person framing, contractions, and conversational connectives that do not appear in the target. The shared topic likely reflects a common course prompt rather than authorial signature.

Stylistic Evidence
  • ›First-person pronouns frequent in sample, absent in target
  • ›Contractions throughout sample, absent in target
  • ›Sentence length divergence (target ≈ 28 words, Student A ≈ 19)
03

Student C

14%

Student C's writing is short, declarative, and morally direct ("They're cheating," "that's not fair"). The target uses none of these features — no short sentences, no moral declaration, no concrete consequence framing. Stylistic profiles diverge across nearly every measured dimension.

Stylistic Evidence
  • ›Mean sentence length divergence (target ≈ 28 words, Student C ≈ 14)
  • ›Student C uses contractions throughout; target uses none
  • ›Student C frames in plain consequence ("didn't do the assignment," "not fair"); target uses abstract institutional framing
Limitations and caveats
  • ⊘Candidate samples are short (~55 words each). Stylometric confidence increases sharply above 300 words per sample.
  • ⊘The target essay is more polished than any of the three candidates' known work. This is consistent with either ghostwriting by a more experienced writer or LLM-assisted editing.
  • ⊘This report does not detect AI authorship. If LLM ghostwriting is suspected, additional methods are required.
  • ⊘All three candidates share a course context, which can pull student writing toward common vocabulary and frames and suppress individual signal.
  • ⊘This report is investigative, not evidentiary. Use as one input alongside human judgment.
This report constitutes investigative analysis only and is not admissible as standalone evidence in any judicial or quasi-judicial proceeding. Stylometric attribution carries irreducible uncertainty even at high confidence. The reader is advised to corroborate findings through other investigative means before acting on them.
Bring us a real case

Five design partners. Free pilot.

Run a case →